This should prevent writing to memory or IO ranges possibly
already allocated for other uses after our device is removed.
v5:
Protect more places wher memcopy_to/form_io takes place
Protect IB submissions
v6: Switch to !drm_dev_enter instead of scoping entire code
with brackets.
v7:
Drop guard of HW ring commands emission protection since they
are in GART and not in MMIO.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210512142648.666476-10-andrey.grodzovsky@amd.com
Problem:
Handle all DMA IOMMU group related dependencies before the
group is removed. Those manifest themself in that when IOMMU
enabled DMA map/unmap is dependent on the presence of IOMMU
group the device belongs to but, this group is released once
the device is removed from PCI topology.
Fix:
Expedite all such unmap operations to pci remove driver callback.
v5: Drop IOMMU notifier and switch to lockless call to ttm_tt_unpopulate
v6: Drop the BO unamp list
v7:
Drop amdgpu_gart_fini
In amdgpu_ih_ring_fini do uncinditional check (!ih->ring)
to avoid freeing uniniitalized rings.
Call amdgpu_ih_ring_fini unconditionally.
v8: Add deatiled explanation
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210517143851.475058-1-andrey.grodzovsky@amd.com
Use it to call disply code dependent on device->drv_data
before it's set to NULL on device unplug
v5:
Move HW finilization into this callback to prevent MMIO accesses
post cpi remove.
v7:
Split kfd suspend from device exit to expdite HW related
stuff to amdgpu_pci_remove
v8:
Squash previous KFD commit into this commit to avoid compile break.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210520032057.497334-1-andrey.grodzovsky@amd.com
Some of the stuff in amdgpu_device_fini such as HW interrupts
disable and pending fences finilization must be done right away on
pci_remove while most of the stuff which relates to finilizing and
releasing driver data structures can be kept until
drm_driver.release hook is called, i.e. when the last device
reference is dropped.
v4: Change functions prefix early->hw and late->sw
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210512142648.666476-3-andrey.grodzovsky@amd.com
mmhub ras is only avaiable in cerntain mmhub ip
generation.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Dennis Li <Dennis.Li@amd.com>
Reviewed-by: John Clements <John.Clements@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
get pf2vf msg info at it's earliest time so that
guest driver can use these info to decide whether
register indirect access enabled.
Signed-off-by: Peng Ju Zhou <PengJu.Zhou@amd.com>
Reviewed-by: Emily.Deng <Emily.Deng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If reset handler is not implemented, reset error before proceeding.
Fixes issue with the following trace -
[ 106.508592] amdgpu 0000:b1:00.0: amdgpu: ASIC reset failed with error, -38 for drm dev, 0000:b1:00.0
[ 106.508972] amdgpu 0000:b1:00.0: amdgpu: GPU reset succeeded, trying to resume
[ 106.509116] [drm] PCIE GART of 512M enabled.
[ 106.509120] [drm] PTB located at 0x0000008000000000
[ 106.509136] [drm] VRAM is lost due to GPU reset!
[ 106.509332] [drm] PSP is resuming...
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-and-tested-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add aldebaran to devices which support recovery
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Expose PG/CG set states functions for other clients
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This prefers reset control based handling if it's implemented
for a particular ASIC. If not, it takes the legacy path. It uses
the legacy method of preparing environment (job, scheduler tasks)
and restoring environment.
v2: remove unused variable (Alex)
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Previous tdr design treats the first job in job_timeout as the bad job.
But sometimes a later bad compute job can block a good gfx job and
cause an unexpected gfx job timeout because gfx and compute ring share
internal GC HW mutually.
[How]
This patch implements an advanced tdr mode.It involves an additinal
synchronous pre-resubmit step(Step0 Resubmit) before normal resubmit
step in order to find the real bad job.
1. At Step0 Resubmit stage, it synchronously submits and pends for the
first job being signaled. If it gets timeout, we identify it as guilty
and do hw reset. After that, we would do the normal resubmit step to
resubmit left jobs.
2. For whole gpu reset(vram lost), do resubmit as the old way.
v2: squash in build fix (Alex)
Signed-off-by: Jack Zhang <Jack.Zhang1@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the following coccicheck warning:
drivers/gpu//drm/amd/amdgpu/amdgpu_ras.c:434:9-17: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_xgmi.c:220:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_xgmi.c:249:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/df_v3_6.c:208:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_psp.c:2973:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_vram_mgr.c:75:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_vram_mgr.c:112:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_vram_mgr.c:58:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_vram_mgr.c:93:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_vram_mgr.c:125:9-17: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_gtt_mgr.c:52:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_gtt_mgr.c:71:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_device.c:140:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_device.c:164:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_device.c:186:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_device.c:208:8-16: WARNING:
use scnprintf or sprintf
drivers/gpu//drm/amd/amdgpu/amdgpu_atombios.c:1916:8-16: WARNING:
use scnprintf or sprintf
Signed-off-by: Tian Tao <tiantao6@hisilicon.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[what]
currently driver recover vram after full access, which may hit
a corner case that meanwhile another whole gpu reset may be
triggered by another VF, which will cause vram recover fail
then fail the whole device reset.
[how]
move the recover vram into full access. So another bad VF will
not disturb the recover sequence for this vf.
Signed-off-by: Horace Chen <horace.chen@amd.com>
Reviewed by: Monk.Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GFX is in gfxoff mode during s0ix so we shouldn't need to
actually tear anything down and restore it.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We handle it properly within the CG/PG functions directly
now.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Not needed as the device is in gfxoff state so the CG/PG state
is handled just like it would be for gfxoff during runtime gfxoff.
This should also prevent delays on resume.
Reworked from Pratik's original patch (Alex)
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Pratik Vishwakarma <Pratik.Vishwakarma@amd.com>
Provide and explanation as to why we skip GFX and PSP for
S0ix. GFX goes into gfxoff, same as runtime, so no need
to tear down and re-init. PSP is part of the always on
state, so no need to touch it.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The SMU expects CGPG to be enabled when entering S0ix.
with this we can re-enable SMU suspend.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This really needs to be done to properly tear down
the device. SMC, PSP, and GFX are still problematic,
need to dig deeper into what aspect of them that is
problematic.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
No functional change.
v2: use correct dev
v3: rework
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move the non-DC specific code into the DCE IP blocks similar
to how we handle DC. This cleans up the common suspend
and resume pathes.
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Set flags at the top level pmops callbacks to track
state. This cleans up the current set of flags and
properly handles S4 on S0ix capable systems.
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During system hibernation suspend still need un-gate gfx CG/PG firstly to handle HW
status check before HW resource destory.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to keep vgaswitcheroo around for HG
systems. They don't use muxes and their power control
is handled via ACPI.
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit 33cf440d59.
And it will enable mode-2 gpu reset for vangogh,
it asks PSP firmware version is 00.1A.00.0F or newer.
Signed-off-by: Xiaojian Du <Xiaojian.Du@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When recovery thread has begun GPU reset, there should be not other
threads to access hardware, otherwise system randomly hang.
v2 (chk): rewritten from scratch, use trylock and lockdep instead of
hand wiring the logic.
v3: add in_irq check
v4: change to check in_task
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We set the same variable a few lines above. Drop the duplicate
setting.
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is to fix the case where it only enable the light SMU
on normal device init. This feature actually need to be enabled after ASIC
been reset as well.
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It was leftover from radeon where it was required for some
specific old hardware. It hasn't been required for ages
and the driver already falls back to MMIO when legacy IO
is not available. Legacy IO also seems to be problematic on
on some thunderbolt devices. Drop it.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: Nicholas Johnson <nicholas.johnson-opensource@outlook.com.au>
Interrupts on are non-reentrant on linux. This is just an ancient
leftover from radeon where irq processing was kicked of from different
places.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
cause:
It is necessary to send ras disable command to ras-ta during gfx
block ras later init, because the ras capability is disable read
from vbios for vega20 gaming, but the ras context is released
during ras init process, this will cause send ras disable command
to ras-to failed.
how:
Delay releasing ras context, the ras context
will be released after gfx block later init done.
Changed from V1:
move release_ras_context into ras_resume
Changed from V2:
check BIT(UMC) is more reasonable before access eeprom table
Signed-off-by: Stanley.Yang <Stanley.Yang@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There is a spelling mistake in a drm debug message. Fix it.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SMU introduce the new interface to enable light Secondary Bus Reset mode, driver
enable it on passthrough + XGMI configuration
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In passthrough configuration, hypervisior will trigger the SBR(Secondary bus reset) to the devices
without sync to each other. This could cause device hang since for XGMI configuration, all the devices
within the hive need to be reset at a limit time slot. This serial of patches try to solve this issue
by co-operate with new SMU which will only do minimum house keeping to response the SBR request but don't
do the real reset job and leave it to driver. Driver need to do the whole sw init and minimum HW init
to bring up the SMU and trigger the reset(possibly BACO) on all the ASICs at the same time
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Acked-by: Andrey Grodzovsky andrey.grodzovsky@amd.com
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The gmc.xgmi.head list originally is designed for device list in the XGMI hive. Mix use it
for reset purpose will prevent the reset function to adjust XGMI device list which is required
in next change
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Andrey Grodzovsky andrey.grodzovsky@amd.com
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu driver may be in reset state during init which will not initialize the kfd,
driver need to initialize the KFD after reset by check the flag
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
driver should use the gfx_info atomfirmware interface
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Feifei Xu <Feifei.Xu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use MSG_GfxDriverReset for mode reset and retire MSG_Mode1Reset.
Centralize soc15_asic_mode1_reset() and nv_asic_mode1_reset()functions.
Add mode2_reset_is_support() for smu->ppt_funcs.
Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Set ip blocks and asic family id
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Parses asic configurations stored in gpu_info firmware and make them available
for driver to use.
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add aldebaran in amdgpu_asic_name array and amdgpu_asic_type enum
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When unloading driver after killing some applications, it will hit sdma
flush tlb job timeout which is called by ttm_bo_delay_delete. So
to avoid the job submit after fence driver fini, call ttm_bo_lock_delayed_workqueue
before fence driver fini. And also put drm_sched_fini before waiting fence.
Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
when try to shutdown guest vm in sriov mode, virt data
exchange is not fini. After vram lost, trying to write
vram could hang cpu.
[How]
add fini virt data exchange in ip_suspend
Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Reviewed-by: Jack Zhang <Jack.Zhang1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
sienna cichlid needs one vf mode which allows vf to set and get
clock status from guest vm. So now expose the required interface
and allow some smu request on VF mode. Also since this asic blocked
direct MMIO access, use KIQ to send SMU request under sriov vf.
OD use same command as getting pp table which is not allowed for
sienna cichlid, so remove OD feature under sriov vf.
Signed-off-by: Horace Chen <horace.chen@amd.com>
Reviewed-by: Monk Liu<monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If the number of badpage records exceed the threshold, driver has
updated both epprom header and control->tbl_hdr.header before gpu reset,
therefore GPU recovery thread no need to read epprom header directly.
v2: merge amdgpu_ras_check_err_threshold into amdgpu_ras_eeprom_check_err_threshold
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move gpu_reset_counter after drm_sched_stop to avoid race
condition caused by job submitted between reset_count +1 and
drm_sched_stop.
Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In the shutdown and poweroff opt on the s0i3 system we still need
un-gate the gfx clock gating and power gating before destory amdgpu device.
Fixes: 628c36d7b2 ("drm/amdgpu: update amdgpu device suspend/resume sequence for s0i3 support")
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1499
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
As dimgrey_cavefish driver is stable enough, set gpu recovery as default
in HW hang for dimgrey_cavefish.
Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Jiansong Chen <Jiansong.Chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Andrey Grodzovsky